reviewate 0.3.0

AI-powered code reviewer for GitHub and GitLab

Reviewate - AI-Powered Code Review

Reviewate is a multi-agent code review system for GitHub and GitLab pull requests. It explores your codebase, verifies every finding against actual code, and posts only fact-checked feedback.

Quick Start

Run without installing (recommended)

uvx reviewate --repo facebook/react --pr 28000 --platform github --review
# First run: prompts for your LLM provider + API key
# Saves config to ~/.reviewate/config.toml — all future runs just work

Requires uv.

Or install globally

uv tool install reviewate   # or: pip install reviewate
reviewate --repo facebook/react --pr 28000 --platform github --review

No token needed if the gh or glab CLI is logged in.

From source (for development)

git clone https://github.com/numberly/reviewate.git
cd reviewate/code_reviewer
uv sync

uv run python main.py --repo facebook/react --pr 28000 --platform github --review

Config file

On first run, Reviewate saves your config to ~/.reviewate/config.toml:

[models]
review = "gemini/gemini-2.5-flash"
utility = "gemini/gemini-2.5-flash-lite"

API keys are read from environment variables only (not stored in the config file).

You can also set env vars or pass CLI args — they take priority over the config file.

Authentication

Method	Best for	Setup
CLI Auth	Local development	Log in with gh auth login or glab auth login
API Token	CI pipelines, Docker, self-hosted	Set GITHUB_API_TOKEN or GITLAB_API_TOKEN

CLI Options

reviewate \
  --repo owner/repo \
  --pr 123 \
  --platform github \
  --review \            # Run review pipeline
  --summary \           # Generate PR summary
  --dry-run \           # Don't post comments
  --review-model anthropic/claude-sonnet-4  # Override review tier model
  --utility-model gemini/gemini-2.5-flash-lite  # Override utility tier model

Architecture

Multi-Agent Pipeline

PR Diff
  │
  ▼
Triage ──▶ 2 Review Agents (parallel, with grep/read tools)
                │
                ▼
           Synthesizer ──▶ Deduplicate ──▶ Fact Checker ──▶ Style
                                                              │
                                                              ▼
                                                        Post Comments

1. TriageAgent — Analyzes PR complexity, determines reasoning effort
2. IssueExplorerAgent — Finds and summarizes linked issues
3. SimpleReviewAgent (x2, parallel) — Reviews code with search/read tools
4. SynthesizerAgent — Combines findings from both reviewers
5. DuplicateAgent — Removes duplicates with existing human comments
6. FactCheckerAgent — Verifies every claim against actual code
7. StyleAgent — Final formatting (makes reviews concise)

Two-Tier Model Configuration

Agents are grouped into tiers so you can use a strong model for critical analysis and a fast/cheap model for supporting tasks:

Tier	Agents	Purpose
REVIEW	triage, review (x2), fact_checker	Critical analysis
UTILITY	style, duplicate, issue_explorer, summarizer, output_parser	Supporting tasks

Code Structure

code_reviewer/
├── main.py              # CLI entry point
├── config.py            # Two-tier model configuration
├── config_file.py       # ~/.reviewate/config.toml handling
├── output.py            # Progress tracking & result output
├── adaptors/
│   ├── factory.py       # Platform handler factory
│   ├── _transport.py    # Abstract transport (APIClient protocol)
│   ├── _http.py         # HTTP client (httpx)
│   ├── _cli.py          # CLI fallback (gh api / glab api)
│   ├── repository/      # PR/MR diffs, reviews, comments
│   └── issue/           # Linked issue fetching
├── agents/
│   ├── base.py          # BaseAgentImpl (LLM calls, tool loop, retry)
│   ├── specialized.py   # SimpleReviewAgent, SynthesizerAgent
│   ├── triage.py        # TriageAgent
│   ├── fact_checker.py  # FactCheckerAgent
│   ├── issue_explorer.py
│   ├── duplicate.py     # DuplicateAgent
│   ├── style.py         # StyleAgent
│   ├── summarizer.py    # SummarizerAgent
│   └── tools.py         # Tool definitions for agents
├── explorer/
│   ├── explorer.py      # CodeExplorer (ripgrep search + file reads)
│   └── builder.py       # Explorer factory
├── workflows/
│   ├── review.py        # ReviewWorkflow (main pipeline)
│   ├── summary.py       # SummaryWorkflow (PR summaries)
│   └── context.py       # Linked repos, team guidelines
├── prompts/             # Jinja2 prompt templates (.txt)
└── schema/              # Data models

Environment Variables

All env vars are optional if you have a config file (~/.reviewate/config.toml). Env vars override config file values when set.

Variable	Description
REVIEW_AGENT_MODEL	Model for review tier agents
REVIEW_AGENT_PROVIDER	Provider for review tier
UTILITY_AGENT_MODEL	Model for utility tier agents
UTILITY_AGENT_PROVIDER	Provider for utility tier
GEMINI_API_KEY	Google Gemini API key
ANTHROPIC_API_KEY	Anthropic Claude API key
OPENAI_API_KEY	OpenAI API key
GITHUB_API_TOKEN	GitHub token (optional with gh CLI)
GITLAB_API_TOKEN	GitLab token (optional with glab CLI)

Override priority: CLI args > env vars > config file

Development

# From monorepo root
make code-review-test  # Run tests
make qa                # Run linters and type checks

License

AGPL-3.0 — see LICENSE file for details.